Dear SPM users<br><br>I&#39;m wondering if there might be a minor mistake in the calculation of the KL-divergence for two gamma densities in function spm_kl_gamma in line 18:<br><div style="margin-left: 40px;">d=d+b_q*c_q/b_p;<br>
</div>with<br><div style="margin-left: 40px;">
KL (Q||P) = &lt;log Q/P&gt; where avg is wrt Q<br>
b_q, c_q    Parameters of first Gamma density<br>b_p, c_p    Parameters of second Gamma density<br></div><br>If I&#39;m correct I would say that line18 should be changed to:<br>
<div style="margin-left: 40px;">d=d+b_p*c_q/b_q;<br></div><br>see e.g. Matthew Beal&#39;s PhD thesis Appendix A (<a href="http://www.cse.buffalo.edu/faculty/mbeal/papers/beal03.pdf" target="_blank">http://www.cse.buffalo.edu/faculty/mbeal/papers/beal03.pdf</a>).<br>


<br>Please let me know what you think. I hope you can help me clarify my uncertainty about the calculation.<br><br>Best regards,<br>Carsten<br clear="all"><br>-- <br>Carsten Stahlhut, Ph.d.-student<br>Informatics and Mathematical Modelling<br>


Richard Petersens Plads, B321/128<br>Technical University of Denmark<br>